This paper presents the application of a Methodology to Obtain Genetic fuzzy rule-based systems Under the iterative rule Learning approach (MOGUL) to forecast energy consumption. Historical data referring to the energy consumption gathered from three groups, namely lights, HVAC and electrical socket, are used to train the proposed approach and achieve forecasting results for the future. The performance of the proposed method is compared to that of previous approaches, namely Hybrid Neural Fuzzy Interface System (HyFIS) and Wang and Mendel’s Fuzzy Rule Learning Method (WM). Results show that the proposed methodology achieved smaller forecasting errors for the following hours, with a smaller standard deviation. Thus, the proposed approach is able to achieve more reliable results than the other state of the art methodologies.

This paper is an extension of work originally presented in 2017

Power and energy systems are being subject to a significant increase of uncertainty. This stochasticity is usually associated to the fast increase of the penetration of renewable generation sources, of intermittent nature, due to their dependency on natural conditions, such as solar intensity and wind speed. However. the uncertainty also plays a relevant role in the consumers side. The variability of consumption is associated to numerous factors, such as consumers’ habits, the environmental temperature, luminosity, energy prices, etc. This makes electricity consumption forecasting crucial to enable dealing with the new paradigm of consumers’ active participation in the power and energy system (

Time series forecasting is an attractive domain in the power and energy systems field, as it is essential to enable and adequate energy resources management. With the increase of renewable generation with an intermittent nature, and the consequent need for the increase in consumers’ flexibility, forecasting energy generation and consumption, in addition to other factors, such as market prices, environmental variables, among many others, becomes crucial (

Several studies have been published related to energy consumption forecast. An approach based on data mining to forecast electricity consumption of a region, based on the meteorological conditions has been presented in (

In (

Also, several studies refer other forecasting methods to forecast the electricity consumption. The study in (

This paper proposes the use of a Methodology to Obtain Genetic fuzzy rule-based systems Under the iterative rule Learning approach (MOGUL) methodology (GFS.FR.MOGUL) to forecast energy consumption of an office building. The objective of this study is to forecast a better profile of the energy consumption for the following hours. Results from the proposed approach are compared to those achieved in previous studies, using different techniques, namely two fuzzy based systems: A Hybrid Neural Fuzzy Inference System (HyFIS) (

After this introductory section, section 2 presents the formulation and explanation of the proposed approach, section 3 presents the achieved results and discusses their comparison to the results achieved by previous methods. Finally, section 4 presents the most relevant conclusions and contributions of this work.

This paper implements a genetic fuzzy system for fuzzy rule learning based on the MOGUL methodology (GFS.FR.MOGUL) to forecast the electricity consumption of an office building. The electricity consumption from building N of the Research Group on Intelligent Engineering and Computing for Advanced Innovation and Development (GECAD) research center located in ISEP/IPP, Porto, Portugal has been chosen to be used in this work. These data is collected and stored through SOICAM (SCADA Office Intelligent Context Awareness Management) (

Genetic fuzzy systems for fuzzy rule learning based on the MOGUL methodology (GFS.FR.MOGUL) is a forecasting method that implements a genetic algorithm determining the structure of the fuzzy IF-THEN rules and the membership function parameters. Two general types of fuzzy IF-THEN rules are considered:

Descriptive rules;

Approximate/free semantic approaches.

In the first type the linguistic labels illustrate a real-world semantic and the linguistic labels are uniformly defined for all rules. In contrast, in the approximate approach there is any associated linguistic label.

Modeling a fuzzy IF-THEN rule on a chromosome which consists of the parameter values of the membership function shows that every rule has the own membership function values. A population contains many such generated chromosomes, based on the iterative rule learning approach (IRL). IRL means that the chromosomes will be generated one by one, taking into account the fitness value and covering factor, until there are sufficient chromosomes in the population. After having obtained the population, the genetic algorithm is started, using the genetic operators’ selection, mutation, and crossover (

Many important statistical properties have to be considered by the Fuzzy Rule Base (FRB) in order to obtain an FRB System FRBS presenting good behavior (Dimiter Driankov, Hans Hellendoorn, Michael Reinfrank, L. Ljung, R. Palm, B. Graham, 1996). In the Generated Fuzzy Rule Bases (GFRB) obtained from MOGUL, there will be consider the satisfaction of two of these statistical properties, which are:

Completeness

Consistency

As an inductive approach to building GFRBSs is considered, both properties will be based on the existence of a training data set, Ep, composed of p numerical input-output problem variable pairs. These examples will present the following structure (1):

As it is explained in (

Where * is a t-norm, and Ri() is the compatibility degree between the rule Ri and the example. Given an FRB composed of T fuzzy rules Ri, the covering value of an example is defined as (5):

And the following condition (6) is required:

A good FRB must satisfy both the conditions presented above, to verify the τ-completeness property and to achieve an appropriate final covering value.

A generic set of IF-THEN rules is consistent if it does not contain contradictions. It is necessary to relax the consistency property to consider the fuzzy rule bases. And it will be done by means of the positive and negative examples concepts (

Positive:

Negative:

and by giving the value k [0,1] and equations (9) and (10):

we get that:

So, the way to incorporate the satisfaction of this property in the designed GFRBSs is to encourage the generation of k-consistent rules. Those rules not verifying this property will be penalized so as not to allow them to be in the FRB finally generated.

In this work the GFS.FR.MOGUL method is implemented in R programming language using FRBS package (

This study uses the SQL server of GECAD research center located in ISEP/IPP, Porto, Portugal. This server includes several databases with the electricity information of GECAD campus buildings. The database of building N of this research center has been used in this work. This building has five energy meters, and each of them stores the electricity consumption data of one specific part of the building with 10 seconds time interval.

A java based application has been developed in this implementation. Which collect the data from the SQL server and calculate the average of the total electricity consumption of the building N – ISEP/GECAD per each hour. This application also creates a new .csv file in a format that can be used as the input of the forecast method.

This work studies on a forecasting approach in order to create a more reliable profile for the electricity consumption in the following hour. In this order, the GFS.FR.MOGUL forecasting method has been chosen to predict the electricity consumption value. The value of the electricity consumption of the intended place during the past 10 days is used to train the GFS.FR.MOGUL method in order to predict the consumption of the next hour.

A java programming language based application has been developed to collect the data from the GECAD SQL server database. The building has 5 energy meters for different parts. The java application receives the recorded consumption of every energy meter and calculates the total energy consumption of the building as well as the average consumption for every hour. The Figure 1 presents a brief perspective of the data sources and data collection of this work.

For every forecasted value the GFS.FR.MOGUL method receives a set of three tables as the input. which are: (i) Testing input, (ii) Training input, and (iii) Training output. The testing input table is the main input of the methods and includes the average of total energy consumption of the building for each hour from last 24 hours before the forecast hour.

The other two tables are used in order to train the GFS.FR.MOGUL method and generate the FUZZY rules. The training input table has the same set of the data as the testing input table but from the last 10 days before the forecasting day. Also, the training output table includes the energy consumption values of the same hour of the target hour from last 10 days. By these two training tables, to forecast every value the GFS.FR.MOGUL method will be trained 10 times by the data from the last 10 days before the target day.

According to equation (1) in this case p is 10 and every set of data from training input table correspond a set of where every value from training output table is the .

In order to test this method, the energy consumption of a total of 12 hours, namely from 12:00 until 23:00 of 16/3/2016, have been forecasted. These hours are chosen because they refer to the hours of greatest activity in the building (higher variation in consumption), thus being the most interesting hours to be forecasted. The MAPE (Mean Absolute Percent Error) error calculation formula is used to compare the forecasted values and the real values of each hour.

During these 12 hours, the largest error belongs to the hour 20 by 16.55% and the lowest error is 0.12% for the hour 13. And also, the average error of these forecasted values is 9.54%.

Many works have been published about forecasting the electricity consumption based on fuzzy rules methods. The work presented in (

As

As the ^{²} of laboratories out of which 3600 m^{²} are clean rooms, and the rest is mainly composed of offices and meeting rooms.

In the (

As it is visible in the

This paper addresses the use of a genetic fuzzy systems for fuzzy rule learning based on the MOGUL methodology to forecast the electricity consumption of an office building in the following hours. This method uses the value of the electricity consumption from last hours to preview this value for the next hour.

By comparing the results of the GFS.FR.MOGUL method to the results of some other previous works it is concluded that this method can present a better electricity consumption profile for the following hours. The presented results in this paper in comparison to the results of some other fuzzy rule based methods, namely HyFIS and WM, are more reliable and all calculated errors ate closer to the average error. In other methods, there is usually a large deference between some calculated errors and the average error of the method. These results show that the GFS.FR.MOGUL method by the average MAPE error of 9.54% is more trustable than the other fuzzy rule base methods. Also, the standard deviation of the forecasting errors of these three methods shows that, the GFS.FR.MOGUL by 5.95% has the lowest standard deviation. It proves the results of this method are more stable and the possibility of having a large error for an hour is less that the other methods. In addition, these results have been compared to the results of a study which also uses the fuzzy rule based methods to predict the energy consumption of a different building. In both cases the GFS.FR.MOGUL provides a more trustable profile for the electricity consumption. The results of these two studies proves that the forecasting methods are able to predict a more reliable value when the electricity consumption value of the intended place is more stable during different hours of the day.

As future work forecasting the electricity consumption for the longer terms such as next days and weeks will be considered as well as studies about the influence of using other variables namely as temperature and humidity on the accuracy of the forecasting the electricity consumption by fuzzy rule based methods.

This work has been developed under the European Union’s Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement No 641794 (project DREAM-GO).